AI, Defense

Meet John Dulin: The Founder Harnessing AI to Keep America Safe

by Isaac Hasson ( 7 min read )
Feb 17, 2023

Modern Intelligence is a seed-stage company applying cutting-edge artificial intelligence to maritime domain awareness. At present, it is challenging for military operators to make sense of all the incoming data on thousands of ships to derive actionable insights on them; Modern Intelligence seeks to address this complex sensor fusion problem through its new product, Cutlass, which it has recently proven at several military exercises. The company has raised $9M and is currently hiring across all functions.

Isaac Hasson: Tell us about yourself, your journey, and how you got interested in defense in the first place.

John Dulin: I originally grew up in a small town in Ohio. I came from a physics background, but after school knew I wanted to work near the intersection of math and deep tech, so I spent time as an early engineer at Freenome, working on machine learning for cancer diagnostics and in quant finance at Numerai (an AI-run hedge fund). I have always been fascinated by defense, but I did not know how to make a difference.

Eventually, I saw that we had a once-in-a-generation opportunity to technically and culturally change defense. The DoD now knows how important AI is, and they are trying to work with startups to speed up their own R&D. A lot of the cynicism regarding acquisitions I originally had was disproven by companies like Anduril, Palantir, Shield, and others.

The vision of Modern coalesced when I joined forces with my co-founder Tristan Tager who was thinking about many of these same themes. Part of Modern’s identity is our unique take on AI in defense. I think everyone agrees the future of defense is AI. However, we also believe the future of AI is defense. As a physicist and history nerd, you cannot help but notice that so many of the significant technological breakthroughs in history came from defense. It was clear that one of the most exciting things to work on was building a defining AI defense and research company.

Isaac Hasson: Tell us about the product you’re building and how you settled on a big problem within defense that AI could address.

John Dulin: A fundamental problem in defense today is detecting, analyzing, and tracking the bad guys instantaneously — turning raw sensor data from drones, satellites, and ships into dots on a map. And there are too many sensors now for humans to watch the screens and do the tracking.

Modern is building multimodal foundation models for perception across every platform. There are countless sensors — most made by contractors who aren’t good at AI — and they all need their target detection, recognition, and tracking automated. But you want the best and most interoperable, so that’ll be one great multimodal AI that does that everywhere, which means being a world-class AI research team.

The DoD needs a cutting-edge AI company with a Top Secret clearance. We are building that by starting with perception, then inference, then decision aids, and finally strategic thinking.

Isaac Hasson: it is notoriously difficult for defense startups to move beyond small R&D contracts and into production, especially at the early stage; I’m curious how you’ve been building a strategy to meet this challenge.

John Dulin: Coming up with a plan to cross “The Valley of Death” is one of the most critical steps in starting a defense company. We wanted to test our maritime AI with the US Southern Command’s drug interdiction mission and the SEALs. We recognized we were outsiders — so we reached out to insiders who knew the problems and players and asked for their help. We built a board of advisors, including former Undersecretary of Defense for Acquisitions and Textron CEO Ellen Lord and former Pentagon Joint AI Center Deputy Director Jacqueline Tame at the strategic level. But even at the product and sales level, it includes our Head of Product, Jerrad Ackerman, who previously led SOCOM strategy at Anduril, a former member of a Naval Special Warfare Team.

The team is tasked with doing a few things all at once. First, the team must understand the warfighter’s needs, which in this case means sorting good ships from bad ships. Second, the team must understand all the stepping stones necessary to scale a defense business. What will our first contract look like? Who are those stakeholders? What will the contract look like five years from now, when we need to have the entire chain of command, joint force, and Congressional Hill bought into our work? And how do the relationships we are building today pave the way for relationships in the future?

Isaac Hasson: We’re seeing a massive influx of excitement in defense, and I think it’s a confluence of a couple of trends. First, as Palmer Luckey noted, until recently, there have been more mattress unicorns than defense unicorns in the past 35 years, but the emergence of companies like Anduril is changing that. That said, Ukraine and great power competition with China have focused people on the consequences of getting this wrong. Where do you think this is all going?

John Dulin: A lot of the great talent of America will return to defense. Both private industry and the defense industry have been coasting. The American military that beat Saddam twice was mostly built in the ’80s, and platforms have been updated on the margin since. At the same time, private industry’s information revolution has been building off defense innovations from the Cold War. A MIT graduate in 1960 might have gone to work at Grumman, DARPA or Skunk Works, but the top talent of the last generation has largely pursued careers in finance or software.

That’s changing now. Partly because we exhausted a lot of the commercialization of that defense R&D from the ’80s, but also because we need to. We need it for great power competition and deterrence with China, and the Ukrainians need it more than anyone.

Isaac Hasson: What advice would you give to people who want to participate as founders, operators, AI researchers, et cetera, but are newcomers to defense specifically?

John Dulin: My advice is to figure out what you’re good at and how that aligns with a defense problem. Tech people in defense can have hubris — they believe they can build anything because they can raise capital. With that approach, you will burn through your funding without breaking through the political ossification that makes acquisitions work.

At Modern, we’re great at AI. If you know what you’re good at, you can create a clear value proposition for a defense user. You can be clear on the skills and network you bring from tech and get them excited about sharing their skills and networks in defense with you. You need those people to build a coalition.

Isaac Hasson: What is your vision for AI in defense, and how does Modern Intelligence fit into it?

John Dulin: If you pick any AI problem in the commercial sector, and then you pick it up, and you put it in defense, it gets way more challenging. For example, take autonomous vehicles like Waymo. The technology understands how to track other cars and objects to recreate human decision-making. This problem is only partially solved despite tens of billions of dollars poured into it. In defense, it’s even harder.

In the commercial sector, you’re controlling a single car. In defense, you’re managing a large constellation of assets, which can include ten drones in one swarm, a satellite, and a boat, all at the same time.

This solution must also help decide which is the best asset to carry out what task in a way that you don’t need to in self-driving cars. Oh — and by the way — the problem is also adversarial because while this is happening, the bad guys are trying to destroy, jam, and evade you.

You need the models to deal with imperfect information that changes in real-time, potentially in communications-denied environments, and you need to do all of that with the highest reliability standards- much higher than anyone else because the lives of innocent people and our warfighters are on the line.

We think defense is where low data and low compute learning will be rewarded because having one big AI sitting in a data center is not an option. You need to have a lot of compute on the edge of drones, tanks, and satellites across a battlefield.

Isaac Hasson: Can you take me through the mechanics of how someone starting a defense company should think about getting a program of record and what you learned about that from your own experience?

John Dulin: It’s very hard. It will take us more time to get there, but we are on the right path. The first step is proving that users and the PMs want your capability. The second step is securing enough recurring revenue to grow, improve, and keep the company alive, all while building coalitions across the DoD and Congress — which takes a long time.

An acquisitions researcher named Eric Lofgren at George Mason wrote an excellent paper called “The History of Thought in Defense Acquisitions.” He shows how the concept of a program of record evolved by telling the history of how the PPBE process, planning, programming, budgeting, and execution evolved in the post-war period. But I like to joke that defense acquisitions are messed up because Robert McNamara believed in Taylorism, in the philosophy of a huge top-down management structure.

Isaac Hasson: He was a central planner.

John Dulin: Exactly. He was a central planner and believed he was reigning in bad spending decisions because, previously, the services had an insane amount of independence. So he created this waterfall acquisitions system called PPBE — Programming, Planning, Budgeting, and Execution — where no DoD PM ever buys anything that isn’t defined by a Pentagon requirement, funded and overseen by Congress, and executed by a Program Office. So when defense leaders talk about “Programs of Record,” they are referring to a program that went through this process. It takes years to work through the cycle. Now, there is an attempt to reform this with the Congressional PPBE Reform Commission that was created last year, which has Modern’s adviser Ellen Lord as Vice Chairwoman — but reform will take a while, and it will only be effective if Congress takes the recommendations seriously.

So when we talk about all of the non-program of record ways we fight for contracts as small startups — OTAs, SBIRs, subcontracts, commercial agreements, etc. — we are talking about workarounds to McNamara’s system.

It is a huge coalition-building endeavor. It needs to be treated as such every step of the way. And even though it is hard work, there is often a good reason for that — technology is only helpful if the warfighter wants it, it fits into the larger mission, and it fits Congress’s policy goals.

Isaac Hasson: Ironically, maybe a bigger multi-agent problem than the one you’re solving in the field.

John Dulin: If we can build an AI that solves that multi-agent problem, we will be set.